Currently each time we submit a job we check the job queue size, and the
drained file. With this change we keep these pieces of information in
memory and don't read them from the filesystem each time.

Significant changes include:
  - The drained value can only be properly set by calling the
    appropriate cluster command "gnt-cluster queue drain/undrain" and
    not by removing/creating the file in the job queue directory. Not
    that anybody would have done it in this undocumented way before.
  - We get rid of the soft limit for the job queue, which we haven't
    ever used anyway.

Signed-off-by: Guido Trotter <ultrotter@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>